diff --git a/xla/service/gpu/gpu_compiler.cc b/xla/service/gpu/gpu_compiler.cc
index 285ef1a2f1f3c..51652f3649b10 100755
--- a/xla/service/gpu/gpu_compiler.cc
+++ b/xla/service/gpu/gpu_compiler.cc
@@ -818,6 +818,7 @@ absl::Status RunOptimizationPasses(
     pipeline.AddPass<HloConstantFolding>();
     pipeline.AddPass<ConditionalSimplifier>();
     pipeline.AddPass<RealImagExpander>();
+    pipeline.AddPass<TransposeFolding>(CanFoldTransposeOperandIntoDot);
     pipeline.AddPass<HloCSE>(/*is_layout_sensitive=*/false);
     pipeline.AddPass<HloDCE>();
   }();
@@ -831,7 +832,6 @@ absl::Status RunOptimizationPasses(
     pipeline.AddPass<ConvertMover>();
     pipeline.AddPass<GpuAlgebraicSimplifier>(layout_insensitive_algsimp_opts,
                                              gpu_version);
-    pipeline.AddPass<TransposeFolding>(CanFoldTransposeOperandIntoDot);
   }();
 
   pipeline.AddPass<HloComputationDeduplicator>(